Difference between revisions of "YASMEEN string transformations"

Revision as of 00:58, 27 October 2013

"Yet Another Species Matching Execution ENgine" - common string transformations

Here follows a list of common string transformations involved in the YASMEEN data conversion and matching processes.

Simplification

This is the process of removing all non-letter and unnecessary characters (symbols, digits, multiple spaces, leading / trailing spaces) from a string, convert the result in the ASCII character set and return the uppercase version of such conversion.

Unnecessary character substitutions is achieved by means of simple RegEx replacements whilst the ASCII character set conversion is delegated to the ICU Java libraries. In particular, the transliterator ID actually used during the simplification process is:

Any-Latin; NFD; [:nonspacing mark:] remove; NFC; Latin-ASCII;

Example of string simplification according to these rules are:

Glaucosoma hebraïcum -> GLAUCOSOMA HEBRAICUM
Cælorhynchis melanosagmatus -> CAELORHYNCHIS MELANOSAGMATUS
One "TwO" three!? :) -> ONE TWO THREE
dOuBlE  Sp4C3S -> DOUBLE SPCS
成开冷渔 -> CHENG KAI LENG YU
حاج عبد الله عابدين -> ALHAJ BD ALLH ABDYN

Stemming

This is the process of removing common sequences of characters appearing at the end of genus and species names. It has the purpose to 'equalize' genus / species names that do differ only for these suffixes (e.g. different Latin genders, singular vs. plural Latin etc.).

It is achieved by means of the following RegEx substitutions:

1. (.*)(IG)(ER|RA|ROS|RUM|RUS)$ -> $1$2
2. (.*)(AE|AK|AM|AR|AS|AX|EA|ES|EX|II|IS|IX|NS|OK|ON|OR|OS|OX|UM|US|YS|YX)$ -> $1
3. (.*)(A|E|I|O|U|Y)$ -> $1 

These substitutions are applied in sequence, from #1 to #3. The first substitution that actually produces changes in the data will halt the propagation.

An example set of stemmed versions for dummy input data (including the applied transformations) is:

NIGER -> NIG (#1)
NIGRA -> NIG (#1)

NIGRES -> NIGR (#2)
NIGREA -> NIGR (#2)
NIGRIGERES -> NIGRIGER (#2)

NIGRICA -> NIGRIC (#3)
NIGRICI -> NIGRIC (#3)

Soundex

The soundex is calculated for simplified string data only, thus ensuring that it will always process ASCII letter characters.

It is applied to every word in the simplified string and to the whole simplified string (spaces removed) according to the original soundex algorithm specification, that is by discarding vowels first and duplicates second. Additionally, no upper limit is set for the maximum length of a soundex.

Thus, the soundex of a multiple-word string will be a sequence of soundexes, one per each word plus the soundex of the overall string. If a word appears multiple times in the simplified string, then its soundex is calculated just once.

Examples of this soundex calculation are:

RHOPALAEA NEAPOLITANA -> R140 N1435 R1451435 (R140 for RHOPALAEA, N1435 for NEAPOLITANA and R1451435 for RHOPALAEANEAPOLITANA)
NIGER NIGER -> N260 N26526 (N260 for NIGER and N26526 for NIGERNIGER)

Trigrams